BLAS: Block-level adaptive striping for solid-state drives

摘要: Increasing the degree of parallelism and reducing overhead garbage collection (GC overhead) are two keys to enhancing performance solid-state drives (SSDs). SSDs employ multichannel architectures, a data placement scheme in an SSD determines how striped channels. Without considering access pattern, existing fixed device-level schemes may have either high GC or poor I/O parallelism, resulting degraded performance. In this article, adaptive block-level called BLAS is proposed maximize while simultaneously minimizing overhead. contrast schemes, allows different policies for blocks with patterns. Pages read-intensive scattered over various channels read pages each remaining attempted be gathered same physical block minimize Moreover, policy logical changed dynamically according pattern changes that block. Finally, parallelism-aware write buffer management approach adopted parallelism. Performance results show yields significant improvement response time when compared schemes. particular, outperforms page striping by factors up 8.75 7.41, respectively. achieves low effective adapting workload changes.